Persistent memory for Claude Code.

01The project memory problem

Claude Code is powerful inside a single session — but your project memory gets lost across time. It forgets repo conventions, architecture decisions, coding standards, past review comments, PR decisions, and the patterns you have banned.

AgentPrizm gives Claude Code a governed memory layer so it remembers the decisions and rules that matter, and can show why it recalled them.

02Connect over MCP

Paste one config block into your MCP client and Claude Code has memory — no local subprocess, no new surface to learn.

{
  "mcpServers": {
    "agentprizm-memory": {
      "type": "http",
      "url": "https://agentprizm.com/api/mcp",
      "headers": { "Authorization": "Bearer ap_..." }
    }
  }
}

03What it remembers

• Repo conventions and folder structure
• Architecture decisions and the reasons behind them
• Repeated PR and review feedback
• Banned patterns and deprecated approaches

04Governed, not just stored

Every recall ships with a receipt — which memories matched, why, and what was filtered — plus confidence scores, fact-validity windows, and right-to-forget. Memory you can trust, and delete on demand.